Direct answer
An AI plagiarism check is really two checks in one: a similarity check that compares your text against a database of published work and student papers, and an AI-writing check that estimates how much of your prose looks machine-generated. Turnitin reports these separately — the similarity score is a percentage of matched text, not a plagiarism verdict, and the AI-writing indicator is a percentage of long-form prose flagged as AI-generated with a sentence-level breakdown [1][2]. Neither number is a finding of misconduct on its own; both are signals your instructor interprets.
What Does an AI Plagiarism Check Actually Detect?
The similarity half of the check works by matching strings of your text against a repository of web pages, academic journals, and previously submitted student papers. Turnitin then highlights every matched passage and links it to the source it matched, so the report shows where overlap occurs rather than simply asserting that you cheated [1]. A high similarity percentage can be produced by perfectly legitimate writing — a long block quote, a standard method description, a references list, or a widely used definition — which is why the number is read alongside the highlighted sources rather than in isolation [1]. What the similarity report cannot tell you is intent: it does not know whether a match was cited, paraphrased badly, or copied.
The AI-writing half is a different measurement entirely, and Turnitin runs it as a separate indicator from the similarity score [2]. It evaluates long-form English prose and returns an overall percentage of the text it considers AI-generated, plus a breakdown that marks which sentences drove that estimate [2]. Because it is a statistical estimate rather than a source match, it does not point at a document you copied from — it points at patterns in your own sentences.
Two practical limits follow from that design. First, the AI indicator only covers prose of sufficient length in English, so short answers, code, equations, and reference lists are outside its scope [2]. Second, Turnitin applies a minimum threshold: when the AI score falls below roughly 20%, the report displays an asterisk (*%) instead of an exact figure, because the signal is low-confidence [2]. A *% is therefore not a clean bill of health and not an accusation — it is Turnitin saying the evidence is too weak to quantify.
How Do You Run an AI Plagiarism Check on Your Own Draft Before You Submit It?
The honest starting point is that students usually cannot self-submit to their institution's Turnitin account. A submission normally has to happen inside a class or assignment that an instructor has enabled, which is why "just check it yourself in Turnitin" is not advice most students can act on [3]. Where an institutional draft option does exist, it typically routes the file into the same student paper repository as a real submission, so a test run can have consequences you did not intend [3].
That repository question is the one worth asking before you upload anything. A non-repository check compares your draft against existing sources without adding your file to the student paper database, which is what you want for a pre-submission preview [3]. You should also check the mechanical limits up front: file formats such as .docx, .pdf, and .txt, a minimum length of a few hundred words, and a file-size ceiling, since a draft that fails those rules returns an error rather than a report [3].
The practical workflow is therefore: finish a full draft, run it through a check that returns both the similarity report and the AI-writing indicator, read the highlighted matches and flagged sentences, then revise before the real submission. Checking a partial draft wastes the run, because both indicators are length-sensitive and a half-finished section gives you a misleadingly low similarity score [3]. Keep the report you generated — it is a dated record of what your draft looked like at that point.
What Should You Do If the Check Flags Your Writing as AI-Generated?
Treat the flag as a prompt to look at specific sentences, not as a verdict. Turnitin's own guidance is that AI-writing indicators require human judgement and are not proof of misconduct, and that formulaic, heavily templated, or heavily edited prose can produce false positives [4]. Open the sentence-level breakdown and read the flagged sentences aloud; the ones that sound generic, over-balanced, or unusually tidy are the ones to rewrite.
Rewrite in your own voice rather than swapping synonyms. Replacing words while keeping the same sentence architecture tends to preserve exactly the statistical patterns that triggered the flag, so the score barely moves [4]. The revision that works is structural: cut the throat-clearing transitions, replace abstract summary sentences with a concrete claim from your reading, add a specific example or number, and vary sentence length so the rhythm is not uniform.
Finally, protect yourself with evidence. Keep dated drafts, version history, notes, and your source PDFs, because a documented writing process is the strongest answer if your work is ever questioned [4]. If your similarity score is the problem rather than the AI score, work through the highlighted matches one by one — quote and cite what should be quoted, and genuinely paraphrase what should not be [1]. And remember that a *% AI result means Turnitin's confidence was too low to report a number, so there is nothing to chase [2].
If you want to see exactly what your institution's system will return — the real similarity report with highlighted sources and the real AI-writing indicator with its sentence breakdown — turnitin0 runs both on your draft in a single check, without adding your file to the student paper database. Students upload a .docx, .pdf, or .txt file over 300 words, and both PDFs come back together, usually in under 15 minutes. It is the same report your professor sees, which makes it a far better guide for revision than guessing.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI plagiarism check the same as an AI detector?
No. Plagiarism checking compares your text to existing sources, while AI detection estimates whether the prose itself was machine-generated; Turnitin reports them as two separate indicators [1][2].
Does a high similarity score mean I plagiarised?
Not necessarily. The score is the percentage of text that matched a source, and correctly cited quotes, references, and standard phrasing all raise it, so the highlighted matches matter more than the number [1].
Can I check my own draft in Turnitin?
Usually not through your institution, because self-submission depends on an instructor enabling it, and institutional drafts may be stored in the student paper repository [3]. A non-repository check gives you the same report without that side effect.
Why does my AI score show an asterisk instead of a percentage?
Turnitin displays *% when the AI result falls below its confidence threshold, which means the signal is too weak to express as a number rather than that the text is clear [2].
Will rewriting a few synonyms fix an AI flag?
Rarely. Synonym swaps keep the same sentence structure and rhythm, so revise the flagged sentences structurally and keep dated drafts as evidence of your process [4].